Can I track the DFT calculation when running the evolutionary algorithm?

[hyesol]

I am currently using "structure-prdiction.toolkit.fixed-composition" workflow to run evolutionary algorithm to search amorphous metal oxide.
Is there a way to track the DFT calculation for each structure?
I want to know the starting energy value, and the energy value for each step of DFT calculation.
I assume I could track it by opening the .zip file, but I was wondering how I can trace the name of .zip file for a specific structure because the name of the .zip file seems like it is generating randomly.
Or is there a better way to do so?

Thank you in advance!

[jacksund]

I would start by checking the directory where your fixed-composition workflow is running. These workflows periodically write some useful output files for you. The most detailed information is in individuals_completed__ALLDATA.csv which you can open up in excel. Sounds like you're after the directory column in this file.

For more complex cases, you can explore the full database tables using a program like DBeaver. This will let you see all the data in an 'excel-like' format. You'd want to look through the relaxation and ionic-step tables to find the information you need.

[hyesol]

Thank you so much for the guidance!
I have come across a situation where the energy values calculated by the evolutionary algorithm seem to differ significantly from the norm/or existing value. For instance, the algorithm calculated an energy value of approximately -10 eV/atom for a CoO structure, but when I did DFT calculations on the same structure, the energy value turned out to be around -6 eV/atom, which is more in line with the expected norm.

I am trying to understand the discrepancy here and whether the initially predicted energy value of -10 eV/atom is reliable. Are there potential sources of error or limitations in the algorithm that could explain such variations in the predicted and calculated values?

[jacksund]

the "algorithm" doesn't calculate the energy of the structure -- it lets DFT do that for it. So you're really comparing DFT values to other DFT values. Specifically the evo search uses relaxation.vasp.staged by default. And which workflow/settings are you running yourself?

I would start thinking about what in a DFT calculation changes it's final energy value. My first thought here is that there are different POTCAR or potentials being used, which in turn will cause a change in the energy. Other issues could be things like running a relaxation in your own calc without running a static energy right after (w. the error being caused by pulay stress).